Claude Code writes it.
Qovery decides what it can ship.
Install eight Qovery skills with one command and Claude Code can deploy applications, provision databases, open preview environments and diagnose failures on your own AWS, GCP, Azure or Scaleway account. The skills are the interface. The permission boundary sits in the API underneath them, where a prompt cannot reach it.
Eight skills, one install.
The Qovery skills follow the Agent Skills open standard, so Claude Code picks the right one from what you ask. Each also has a slash command if you would rather be explicit.
qovery
/qoveryEntry point. Routes to the right skill and handles quick operations: list, status, stop, restart, logs, clone, scale, custom domains, deployment history.
qovery-onboard
/qovery-onboardGuided setup. Recommends cloud provider, cluster type, environment structure, RBAC and cost defaults. Handles bring-your-own-Kubernetes and migrations from Heroku, Vercel, Netlify, Render and Railway.
qovery-deploy
/qovery-deployDeploys any application to Kubernetes. Analyses the codebase, writes a Dockerfile if one is missing, provisions databases, sets environment variables, then deploys via CLI and API or Terraform.
qovery-troubleshoot
/qovery-troubleshootEight-layer diagnosis across deployment status, build and runtime logs, health checks, env vars, networking, resources and cluster. Matches 20+ error patterns and writes a runbook to .qovery/runbooks/.
qovery-optimize
/qovery-optimizeAnalyses historical CPU and memory against allocation across seven optimisation dimensions, estimates external cloud costs, and produces a markdown report plus CSV export.
qovery-speedup
/qovery-speedupMeasures the deployment pipeline stage by stage, classifies each bottleneck as yours or Qovery infrastructure, and proposes Dockerfile and health-check fixes with before and after timings.
qovery-preview
/qovery-previewCreates a preview environment from a pull request. Detects or builds a blueprint environment, clones it, switches branches, and configures auto-shutdown so it does not run all weekend.
qovery-terraform
/qovery-terraformReads an existing Qovery setup from the API and generates HCL for the qovery/qovery provider, imports resources into state, and validates against a test clone before you commit.
Two interfaces. Pick both.
The skills deploy. The MCP Server operates what is already running. They are complementary, and most teams end up using both.
Agent Skills
The skills follow the Agent Skills open standard. The installer places all eight where Claude Code discovers them, in ~/.claude/skills/, and installs the matching slash commands.
Then just ask: "Deploy my application with Qovery".
# Install all eight skills, globally $ curl -fsSL https://skill.qovery.com/install.sh | bash # Or scope them to the current project only $ curl -fsSL https://skill.qovery.com/install.sh | bash -s -- --project
MCP Server
The skills deploy. The MCP Server operates what is already running. Claude Code supports OAuth, so no token generation is needed.
Write mode also requires write access to be enabled in the Qovery Console.
# Read-only (default)
$ claude mcp add --transport http qovery \
https://mcp.qovery.com/mcp --callback-port 4242
# Read/write, when you explicitly want it
$ claude mcp add --transport http qovery \
"https://mcp.qovery.com/mcp?read_write=true" --callback-port 4242The boundary is not in the prompt.
It is in the API underneath.
Instructions in a file are a request. Claude Code can read them, and it can also decide not to follow them. These four properties hold regardless of what the model decides, because none of them are things the model is asked to respect.
Every one of them is checkable in the docs before you trust a word of this page.
Read-only by default
The Qovery MCP Server connects in read-only mode unless you say otherwise. Your agent can list environments, read service state, and tail logs. It cannot change anything.
Write access is two deliberate switches
Writes require read_write=true on the connection URL and a separate write-access setting enabled in the Qovery Console. Neither one is something a model can talk its way into.
Scoped by the RBAC you already run
An agent authenticates with its own API token and inherits the permissions of the role that token was issued under. Agents do not get a parallel access model, and they do not consume user seats.
Every action lands in the audit log
MCP tool calls and API operations are recorded in the Qovery Console audit trail against the token that made them, which is what turns "an agent did something" into an answerable question.
Claude Code and Qovery.
How do I connect Claude Code to Qovery?+
Run curl -fsSL https://skill.qovery.com/install.sh | bash to install the eight Qovery agent skills, then ask Claude Code to deploy your application. To also manage running infrastructure, add the MCP Server with claude mcp add --transport http qovery https://mcp.qovery.com/mcp --callback-port 4242.
Can Claude Code deploy to my own AWS account?+
Yes. Qovery installs into your own AWS, GCP, Azure or Scaleway account, so when Claude Code deploys through the Qovery skills the workload runs on your cluster, in your VPC, under your compliance boundary. Qovery operates the infrastructure; it does not host your applications.
What stops Claude Code from doing something destructive?+
The boundary is outside the model. MCP access is read-only by default and write access requires both a read_write=true connection parameter and a separate setting in the Qovery Console. The agent authenticates with its own API token and inherits that token role permissions, and every action is recorded in the audit log. A prompt cannot change any of this.
Do AI agents consume Qovery user seats?+
No. Agents authenticate with their own API tokens and there is no limit on how many you run. They inherit scoped RBAC permissions and are audited exactly like a human user. What they consume is deployment minutes and environments.
What is the difference between the Qovery skills and the MCP Server?+
The skills deploy and the MCP Server operates. Skills handle going from source code to a running application: analysing the codebase, writing a Dockerfile, provisioning a database, deploying. The MCP Server is for managing infrastructure that already exists: querying state, reading logs, and triggering deployments. Most teams use both.
One command, and Claude Code
deploys to your own cloud.
Install the skills, point them at your AWS, GCP, Azure or Scaleway account, and let the agent work inside a boundary you set once. Read-only until you decide otherwise.